Intel officials continue to aggressively expand the reach of its x86-based processors throughout the data center as they look to offer Intel Architecture chips for every workload and to muscle out any potential challenge in servers and other infrastructure systems from ARM Holdings and others.
Intel last month introduced its Xeon Phi coprocessors, aimed at organizations with high-performance computing (HPC) environments and massive systems that run compute-intensive and highly parallel workloads. Many of these organizations have turned to GPU accelerators from the likes of Nvidia and Advanced Micro Devices to help increase performance while keeping power consumption in check. Intel’s Xeon Phi coprocessors are being offered as an alternative to GPU accelerators.
Now the giant chip maker is looking at the other end of the data center spectrum. Company officials Dec. 11 unveiled its dual-core Atom S1200 products, a family of low-power (6-watt) systems-on-a-chip (SoCs) that are designed for microservers. Microservers are small, highly energy-efficient systems aimed at extremely dense environments such as dedicated Web hosting facilities, which have to move massive amounts of lightweight workloads, and where power consumption is at a premium. The 6 watts is half what other low-power server chips offer in energy consumption, according to Intel officials.
These computing environments also are where, because of the need for low-power consumption, ARM and its partners—such as Samsung Electronics, Nvidia, Calxeda, Marvell Technologies and now AMD—see an opportunity to move their architectures up the ladder, from the mobile-device space and into the data center.
However, Intel executives see a clear advantage in their Atom platform, which offers a host of capabilities—from 64-bit computing and virtualization support to Error Correcting Code (ECC) and greater memory capacity—that are crucial to servers and won’t appear in ARM-designed chips until late 2013 at the latest, and more like 2014. In addition, because Atom is based on the x86 architecture, it comes with familiar software and developers’ tools. It is a similar argument Intel executives use when talking about Xeon Phi—also based on x86—when compared with graphics accelerators.
The Atom S1200 chips—along with the family of Xeon E3, E5 and E7 chips, the Xeon Phi coprocessors, and the company’s networking technologies—give Intel complete coverage of data center workloads, Diane Bryant, vice president and general manager of Intel’s Data Center and Connected Systems Group, said during a Webcast press conference Dec. 11. In addition, the new Atom SoCs also can be used in storage and communications and networking appliances, Bryant said.
“Whatever the workload is, it works best on Intel Architecture,” she said.
Intel Brings 6-Watt Atom Server SoC to the Data Center
Intel during the press conference got backup from executives from Hewlett-Packard, Microsoft and Facebook, both of whom said the new Atom SoCs offer a strong combination of server capabilities with high energy efficiency. HP last year became the first top-tier server OEM to announce it was partnering with a company—Calxeda—developing ARM-based server chips as part of its larger Project Moonshot initiative to develop dense, highly power-efficient microservers.
However, HP officials announced the first of these “Gemini” systems would be based on Intel’s 64-bit Atom “Centerton” chips. During the Webcast, Paul Santeler, vice president of HP’s Hyperscale Business Unit, lauded the Atom S1200 chips, calling the product family “a true server-class product built on a truly energy-efficient core.”
HP already is shipping the first of its Gemini servers, and in the first quarter next year will begin shipping the second generation, Santeler said.
Bryant said Intel already has more than 20 design wins in the works based on the Atom S1200 chips from the likes of HP, Dell, Huawei, SuperMicro and Quanta. Some of those partners have been able to get as many as 1,000 of the Atom SoCs into a single server rack, she said.
Intel is already prepping for the next generation of the Atom server SoCs—called “Avoton”—which will be based on the 22-nanometer manufacturing process and include the company’s 3D Tri-Gate transistor architecture, Jason Waxman, general manager of Intel’s Cloud Platform Group, wrote in a Dec. 10 blog post. In 2014, a 14nm Atom platform will be released. With each new generation, Intel also will integrate other features, such as networking technology, into the chips, Bryant said.
Intel’s announcement was quickly dismissed by AMD executives, who argued that Intel was too late in getting into the microserver space. They noted their $334 million acquisition in February of SeaMicro, a company making microservers that at one time had partnered with Intel. In an email sent to journalists, AMD officials pointed to the SeaMicro deal and its partnership with ARM to develop ARM-based server chips as examples of their company’s leadership in the space.
“Intel is way behind on small cores,” they said in the email. “They have no cell phone market share, little tablet market share, and now they are threatened that they will lose server market share. AMD and its SeaMicro technology are leading the charge in microserver technology and development.”
The microserver market is still very much a niche market, but one that Intel executives have said could grow to as much as 10 percent of the overall server space, while some analysts have said it could be more than 15 to 20 percent.
Intel Brings 6-Watt Atom Server SoC to the Data Center
Charles King, principal analyst with Pund-IT Research, said in a Dec. 11 research note that ARM and its partners deserve credit for their work in defining modern mobile computing with their power-efficient chips. King also acknowledged arguments by ARM proponents that the company’s history of developing low-power chips for such devices as smartphones and tablets give it an edge in the developing microserver space.
However, he said ARM faces some difficult challenges, including that Intel is already in the market with a server-ready, low-power SoC, while ARM and its partners are still more than a year away. Some partners, like Calxeda and Marvell, already sell ARM-based server chips, but they’re 32-bit products.
“Though ARM recently announced its next-generation 64-bit Cortex-A57 and Cortex-A53 processor designs, products based on the new architecture likely won’t be available until late 2013 at the earliest (vendors, including AMD, have already stated that they plan to deliver Cortex-based silicon in 2014),” King wrote. “Since Intel’s Atom road map includes new 22nm products (estimated for 2013) and future 14nm solutions, when the first 64-bit ARM processors finally do arrive in the marketplace, they’ll be facing the second generation of Intel Atom designed specifically for microservers.”
In addition, ARM supporters shouldn’t overlook Intel’s advantage in leveraging the x86 architecture.
“Why is this a big deal? For two reasons. First and foremost, current owners of Intel-compatible applications will be able to utilize those solutions in Atom S1200-based systems today and capture immediate microserver benefits,” King wrote. “In addition, ISVs and other developers interested in exploring the microserver market won’t be forced into porting or rewriting their applications and tools for various new (ARM) platforms.”
However, the Atom SoCs could present some problems for Intel, according to Vijay Rakesh, an analyst with Sterne Agee. In a Dec. 12 research note, Rakesh said the new Atom platform “is a positive [sign] that Intel is recognizing the low-power server market,” but added that at a price of $54 per 1,000 units, it could hurt the chip maker’s margins by possibly undercutting sales of more expensive Xeons.